Aimed at researchers building browser agents who want open weights, training data and code rather than a finished product.
Kenning's take
What it does
MolmoWeb is an open web agent built on Ai2's Molmo 2 models, in 4B and 8B sizes. Given a task and a live web page, it looks at a screenshot, writes a short thought, and chooses the next browser action such as clicking, typing or scrolling. Ai2 released the weights, the MolmoWebMix training data, training code and evaluation tools, and runs a public demo.
- 4B and 8B parameter models built on Molmo 2
- Works from screenshots rather than page code
- Clicks, types, scrolls and switches tabs
- MolmoWebMix training dataset of human and synthetic browsing
- Training code, evaluation harness and annotation tool on GitHub
Who it suits. Researchers and developers building or studying open browser agents.
Key facts
- Category
- AI models
- Made by
- Ai2 (Allen Institute for AI)
- Based in
- United States
- Released
- 24 March 2026
- How to get it
- weights on Hugging Face and code on GitHub for self-hosting; a public web demo at molmoweb.allen.ai
- Works on
- self-hosted, Web
- In the UK
- Yes
- Free plan
- free open weights, data and code
FAQ
What is MolmoWeb?
An open AI agent from Ai2 that controls a web browser by reading screenshots and choosing actions like clicking and typing.
Is MolmoWeb free?
Yes. The weights, training data and code are free to download, and the code is Apache 2.0.
When was MolmoWeb released?
Ai2 announced it on 24 March 2026 and released the full codebase on 10 April 2026.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.